Linux Audio

Check our new training course

Loading...
v4.6
 
  1/*  KVM paravirtual clock driver. A clocksource implementation
  2    Copyright (C) 2008 Glauber de Oliveira Costa, Red Hat Inc.
  3
  4    This program is free software; you can redistribute it and/or modify
  5    it under the terms of the GNU General Public License as published by
  6    the Free Software Foundation; either version 2 of the License, or
  7    (at your option) any later version.
  8
  9    This program is distributed in the hope that it will be useful,
 10    but WITHOUT ANY WARRANTY; without even the implied warranty of
 11    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 12    GNU General Public License for more details.
 13
 14    You should have received a copy of the GNU General Public License
 15    along with this program; if not, write to the Free Software
 16    Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
 17*/
 18
 19#include <linux/clocksource.h>
 20#include <linux/kvm_para.h>
 21#include <asm/pvclock.h>
 22#include <asm/msr.h>
 23#include <asm/apic.h>
 24#include <linux/percpu.h>
 25#include <linux/hardirq.h>
 26#include <linux/memblock.h>
 27#include <linux/sched.h>
 
 
 
 
 28
 
 
 29#include <asm/x86_init.h>
 30#include <asm/reboot.h>
 
 31
 32static int kvmclock = 1;
 33static int msr_kvm_system_time = MSR_KVM_SYSTEM_TIME;
 34static int msr_kvm_wall_clock = MSR_KVM_WALL_CLOCK;
 35static cycle_t kvm_sched_clock_offset;
 
 36
 37static int parse_no_kvmclock(char *arg)
 38{
 39	kvmclock = 0;
 40	return 0;
 41}
 42early_param("no-kvmclock", parse_no_kvmclock);
 43
 44/* The hypervisor will put information about time periodically here */
 45static struct pvclock_vsyscall_time_info *hv_clock;
 46static struct pvclock_wall_clock wall_clock;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 47
 48struct pvclock_vsyscall_time_info *pvclock_pvti_cpu0_va(void)
 49{
 50	return hv_clock;
 
 
 
 
 
 51}
 52
 53/*
 54 * The wallclock is the time of day when we booted. Since then, some time may
 55 * have elapsed since the hypervisor wrote the data. So we try to account for
 56 * that with system time
 57 */
 58static void kvm_get_wallclock(struct timespec *now)
 59{
 60	struct pvclock_vcpu_time_info *vcpu_time;
 61	int low, high;
 62	int cpu;
 63
 64	low = (int)__pa_symbol(&wall_clock);
 65	high = ((u64)__pa_symbol(&wall_clock) >> 32);
 66
 67	native_write_msr(msr_kvm_wall_clock, low, high);
 68
 69	cpu = get_cpu();
 70
 71	vcpu_time = &hv_clock[cpu].pvti;
 72	pvclock_read_wallclock(&wall_clock, vcpu_time, now);
 73
 74	put_cpu();
 75}
 76
 77static int kvm_set_wallclock(const struct timespec *now)
 78{
 79	return -1;
 80}
 81
 82static cycle_t kvm_clock_read(void)
 83{
 84	struct pvclock_vcpu_time_info *src;
 85	cycle_t ret;
 86	int cpu;
 87
 88	preempt_disable_notrace();
 89	cpu = smp_processor_id();
 90	src = &hv_clock[cpu].pvti;
 91	ret = pvclock_clocksource_read(src);
 92	preempt_enable_notrace();
 93	return ret;
 94}
 95
 96static cycle_t kvm_clock_get_cycles(struct clocksource *cs)
 97{
 98	return kvm_clock_read();
 99}
100
101static cycle_t kvm_sched_clock_read(void)
102{
103	return kvm_clock_read() - kvm_sched_clock_offset;
104}
105
106static inline void kvm_sched_clock_init(bool stable)
107{
108	if (!stable) {
109		pv_time_ops.sched_clock = kvm_clock_read;
110		return;
111	}
112
113	kvm_sched_clock_offset = kvm_clock_read();
114	pv_time_ops.sched_clock = kvm_sched_clock_read;
115	set_sched_clock_stable();
116
117	printk(KERN_INFO "kvm-clock: using sched offset of %llu cycles\n",
118			kvm_sched_clock_offset);
119
120	BUILD_BUG_ON(sizeof(kvm_sched_clock_offset) >
121	         sizeof(((struct pvclock_vcpu_time_info *)NULL)->system_time));
122}
123
124/*
125 * If we don't do that, there is the possibility that the guest
126 * will calibrate under heavy load - thus, getting a lower lpj -
127 * and execute the delays themselves without load. This is wrong,
128 * because no delay loop can finish beforehand.
129 * Any heuristics is subject to fail, because ultimately, a large
130 * poll of guests can be running and trouble each other. So we preset
131 * lpj here
132 */
133static unsigned long kvm_get_tsc_khz(void)
134{
135	struct pvclock_vcpu_time_info *src;
136	int cpu;
137	unsigned long tsc_khz;
138
139	cpu = get_cpu();
140	src = &hv_clock[cpu].pvti;
141	tsc_khz = pvclock_tsc_khz(src);
142	put_cpu();
143	return tsc_khz;
144}
145
146static void kvm_get_preset_lpj(void)
147{
148	unsigned long khz;
149	u64 lpj;
150
151	khz = kvm_get_tsc_khz();
152
153	lpj = ((u64)khz * 1000);
154	do_div(lpj, HZ);
155	preset_lpj = lpj;
156}
157
158bool kvm_check_and_clear_guest_paused(void)
159{
 
160	bool ret = false;
161	struct pvclock_vcpu_time_info *src;
162	int cpu = smp_processor_id();
163
164	if (!hv_clock)
165		return ret;
166
167	src = &hv_clock[cpu].pvti;
168	if ((src->flags & PVCLOCK_GUEST_STOPPED) != 0) {
169		src->flags &= ~PVCLOCK_GUEST_STOPPED;
170		pvclock_touch_watchdogs();
171		ret = true;
172	}
173
174	return ret;
175}
176
177static struct clocksource kvm_clock = {
178	.name = "kvm-clock",
179	.read = kvm_clock_get_cycles,
180	.rating = 400,
181	.mask = CLOCKSOURCE_MASK(64),
182	.flags = CLOCK_SOURCE_IS_CONTINUOUS,
 
 
 
 
 
 
 
183};
 
184
185int kvm_register_clock(char *txt)
186{
187	int cpu = smp_processor_id();
188	int low, high, ret;
189	struct pvclock_vcpu_time_info *src;
190
191	if (!hv_clock)
192		return 0;
193
194	src = &hv_clock[cpu].pvti;
195	low = (int)slow_virt_to_phys(src) | 1;
196	high = ((u64)slow_virt_to_phys(src) >> 32);
197	ret = native_write_msr_safe(msr_kvm_system_time, low, high);
198	printk(KERN_INFO "kvm-clock: cpu %d, msr %x:%x, %s\n",
199	       cpu, high, low, txt);
200
201	return ret;
 
 
202}
203
204static void kvm_save_sched_clock_state(void)
205{
206}
207
208static void kvm_restore_sched_clock_state(void)
209{
210	kvm_register_clock("primary cpu clock, resume");
211}
212
213#ifdef CONFIG_X86_LOCAL_APIC
214static void kvm_setup_secondary_clock(void)
215{
216	/*
217	 * Now that the first cpu already had this clocksource initialized,
218	 * we shouldn't fail.
219	 */
220	WARN_ON(kvm_register_clock("secondary cpu clock"));
221}
222#endif
223
224/*
225 * After the clock is registered, the host will keep writing to the
226 * registered memory location. If the guest happens to shutdown, this memory
227 * won't be valid. In cases like kexec, in which you install a new kernel, this
228 * means a random memory location will be kept being written. So before any
229 * kind of shutdown from our side, we unregister the clock by writing anything
230 * that does not have the 'enable' bit set in the msr
231 */
232#ifdef CONFIG_KEXEC_CORE
233static void kvm_crash_shutdown(struct pt_regs *regs)
234{
235	native_write_msr(msr_kvm_system_time, 0, 0);
236	kvm_disable_steal_time();
237	native_machine_crash_shutdown(regs);
238}
239#endif
240
241static void kvm_shutdown(void)
242{
243	native_write_msr(msr_kvm_system_time, 0, 0);
244	kvm_disable_steal_time();
245	native_machine_shutdown();
246}
247
248void __init kvmclock_init(void)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
249{
250	struct pvclock_vcpu_time_info *vcpu_time;
251	unsigned long mem;
252	int size, cpu;
253	u8 flags;
254
255	size = PAGE_ALIGN(sizeof(struct pvclock_vsyscall_time_info)*NR_CPUS);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
256
257	if (!kvm_para_available())
258		return;
259
260	if (kvmclock && kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE2)) {
261		msr_kvm_system_time = MSR_KVM_SYSTEM_TIME_NEW;
262		msr_kvm_wall_clock = MSR_KVM_WALL_CLOCK_NEW;
263	} else if (!(kvmclock && kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE)))
264		return;
265
266	printk(KERN_INFO "kvm-clock: Using msrs %x and %x",
267		msr_kvm_system_time, msr_kvm_wall_clock);
268
269	mem = memblock_alloc(size, PAGE_SIZE);
270	if (!mem)
271		return;
272	hv_clock = __va(mem);
273	memset(hv_clock, 0, size);
274
275	if (kvm_register_clock("primary cpu clock")) {
276		hv_clock = NULL;
277		memblock_free(mem, size);
278		return;
279	}
280
 
 
 
 
 
 
 
281	if (kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE_STABLE_BIT))
282		pvclock_set_flags(PVCLOCK_TSC_STABLE_BIT);
283
284	cpu = get_cpu();
285	vcpu_time = &hv_clock[cpu].pvti;
286	flags = pvclock_read_flags(vcpu_time);
287
288	kvm_sched_clock_init(flags & PVCLOCK_TSC_STABLE_BIT);
289	put_cpu();
290
291	x86_platform.calibrate_tsc = kvm_get_tsc_khz;
 
292	x86_platform.get_wallclock = kvm_get_wallclock;
293	x86_platform.set_wallclock = kvm_set_wallclock;
294#ifdef CONFIG_X86_LOCAL_APIC
295	x86_cpuinit.early_percpu_clock_init =
296		kvm_setup_secondary_clock;
297#endif
298	x86_platform.save_sched_clock_state = kvm_save_sched_clock_state;
299	x86_platform.restore_sched_clock_state = kvm_restore_sched_clock_state;
300	machine_ops.shutdown  = kvm_shutdown;
301#ifdef CONFIG_KEXEC_CORE
302	machine_ops.crash_shutdown  = kvm_crash_shutdown;
303#endif
304	kvm_get_preset_lpj();
305	clocksource_register_hz(&kvm_clock, NSEC_PER_SEC);
306	pv_info.name = "KVM";
307}
308
309int __init kvm_setup_vsyscall_timeinfo(void)
310{
311#ifdef CONFIG_X86_64
312	int cpu;
313	u8 flags;
314	struct pvclock_vcpu_time_info *vcpu_time;
315	unsigned int size;
316
317	if (!hv_clock)
318		return 0;
319
320	size = PAGE_ALIGN(sizeof(struct pvclock_vsyscall_time_info)*NR_CPUS);
321
322	cpu = get_cpu();
323
324	vcpu_time = &hv_clock[cpu].pvti;
325	flags = pvclock_read_flags(vcpu_time);
326
327	if (!(flags & PVCLOCK_TSC_STABLE_BIT)) {
328		put_cpu();
329		return 1;
330	}
331
332	put_cpu();
 
 
 
 
 
 
 
 
 
 
 
333
334	kvm_clock.archdata.vclock_mode = VCLOCK_PVCLOCK;
335#endif
336	return 0;
337}
v5.9
  1// SPDX-License-Identifier: GPL-2.0-or-later
  2/*  KVM paravirtual clock driver. A clocksource implementation
  3    Copyright (C) 2008 Glauber de Oliveira Costa, Red Hat Inc.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  4*/
  5
  6#include <linux/clocksource.h>
  7#include <linux/kvm_para.h>
  8#include <asm/pvclock.h>
  9#include <asm/msr.h>
 10#include <asm/apic.h>
 11#include <linux/percpu.h>
 12#include <linux/hardirq.h>
 13#include <linux/cpuhotplug.h>
 14#include <linux/sched.h>
 15#include <linux/sched/clock.h>
 16#include <linux/mm.h>
 17#include <linux/slab.h>
 18#include <linux/set_memory.h>
 19
 20#include <asm/hypervisor.h>
 21#include <asm/mem_encrypt.h>
 22#include <asm/x86_init.h>
 23#include <asm/reboot.h>
 24#include <asm/kvmclock.h>
 25
 26static int kvmclock __initdata = 1;
 27static int kvmclock_vsyscall __initdata = 1;
 28static int msr_kvm_system_time __ro_after_init = MSR_KVM_SYSTEM_TIME;
 29static int msr_kvm_wall_clock __ro_after_init = MSR_KVM_WALL_CLOCK;
 30static u64 kvm_sched_clock_offset __ro_after_init;
 31
 32static int __init parse_no_kvmclock(char *arg)
 33{
 34	kvmclock = 0;
 35	return 0;
 36}
 37early_param("no-kvmclock", parse_no_kvmclock);
 38
 39static int __init parse_no_kvmclock_vsyscall(char *arg)
 40{
 41	kvmclock_vsyscall = 0;
 42	return 0;
 43}
 44early_param("no-kvmclock-vsyscall", parse_no_kvmclock_vsyscall);
 45
 46/* Aligned to page sizes to match whats mapped via vsyscalls to userspace */
 47#define HV_CLOCK_SIZE	(sizeof(struct pvclock_vsyscall_time_info) * NR_CPUS)
 48#define HVC_BOOT_ARRAY_SIZE \
 49	(PAGE_SIZE / sizeof(struct pvclock_vsyscall_time_info))
 50
 51static struct pvclock_vsyscall_time_info
 52			hv_clock_boot[HVC_BOOT_ARRAY_SIZE] __bss_decrypted __aligned(PAGE_SIZE);
 53static struct pvclock_wall_clock wall_clock __bss_decrypted;
 54static DEFINE_PER_CPU(struct pvclock_vsyscall_time_info *, hv_clock_per_cpu);
 55static struct pvclock_vsyscall_time_info *hvclock_mem;
 56
 57static inline struct pvclock_vcpu_time_info *this_cpu_pvti(void)
 58{
 59	return &this_cpu_read(hv_clock_per_cpu)->pvti;
 60}
 61
 62static inline struct pvclock_vsyscall_time_info *this_cpu_hvclock(void)
 63{
 64	return this_cpu_read(hv_clock_per_cpu);
 65}
 66
 67/*
 68 * The wallclock is the time of day when we booted. Since then, some time may
 69 * have elapsed since the hypervisor wrote the data. So we try to account for
 70 * that with system time
 71 */
 72static void kvm_get_wallclock(struct timespec64 *now)
 73{
 74	wrmsrl(msr_kvm_wall_clock, slow_virt_to_phys(&wall_clock));
 75	preempt_disable();
 76	pvclock_read_wallclock(&wall_clock, this_cpu_pvti(), now);
 77	preempt_enable();
 
 
 
 
 
 
 
 
 
 
 
 78}
 79
 80static int kvm_set_wallclock(const struct timespec64 *now)
 81{
 82	return -ENODEV;
 83}
 84
 85static u64 kvm_clock_read(void)
 86{
 87	u64 ret;
 
 
 88
 89	preempt_disable_notrace();
 90	ret = pvclock_clocksource_read(this_cpu_pvti());
 
 
 91	preempt_enable_notrace();
 92	return ret;
 93}
 94
 95static u64 kvm_clock_get_cycles(struct clocksource *cs)
 96{
 97	return kvm_clock_read();
 98}
 99
100static u64 kvm_sched_clock_read(void)
101{
102	return kvm_clock_read() - kvm_sched_clock_offset;
103}
104
105static inline void kvm_sched_clock_init(bool stable)
106{
107	if (!stable)
108		clear_sched_clock_stable();
 
 
 
109	kvm_sched_clock_offset = kvm_clock_read();
110	pv_ops.time.sched_clock = kvm_sched_clock_read;
 
111
112	pr_info("kvm-clock: using sched offset of %llu cycles",
113		kvm_sched_clock_offset);
114
115	BUILD_BUG_ON(sizeof(kvm_sched_clock_offset) >
116		sizeof(((struct pvclock_vcpu_time_info *)NULL)->system_time));
117}
118
119/*
120 * If we don't do that, there is the possibility that the guest
121 * will calibrate under heavy load - thus, getting a lower lpj -
122 * and execute the delays themselves without load. This is wrong,
123 * because no delay loop can finish beforehand.
124 * Any heuristics is subject to fail, because ultimately, a large
125 * poll of guests can be running and trouble each other. So we preset
126 * lpj here
127 */
128static unsigned long kvm_get_tsc_khz(void)
129{
130	setup_force_cpu_cap(X86_FEATURE_TSC_KNOWN_FREQ);
131	return pvclock_tsc_khz(this_cpu_pvti());
 
 
 
 
 
 
 
132}
133
134static void __init kvm_get_preset_lpj(void)
135{
136	unsigned long khz;
137	u64 lpj;
138
139	khz = kvm_get_tsc_khz();
140
141	lpj = ((u64)khz * 1000);
142	do_div(lpj, HZ);
143	preset_lpj = lpj;
144}
145
146bool kvm_check_and_clear_guest_paused(void)
147{
148	struct pvclock_vsyscall_time_info *src = this_cpu_hvclock();
149	bool ret = false;
 
 
150
151	if (!src)
152		return ret;
153
154	if ((src->pvti.flags & PVCLOCK_GUEST_STOPPED) != 0) {
155		src->pvti.flags &= ~PVCLOCK_GUEST_STOPPED;
 
156		pvclock_touch_watchdogs();
157		ret = true;
158	}
 
159	return ret;
160}
161
162static int kvm_cs_enable(struct clocksource *cs)
163{
164	vclocks_set_used(VDSO_CLOCKMODE_PVCLOCK);
165	return 0;
166}
167
168struct clocksource kvm_clock = {
169	.name	= "kvm-clock",
170	.read	= kvm_clock_get_cycles,
171	.rating	= 400,
172	.mask	= CLOCKSOURCE_MASK(64),
173	.flags	= CLOCK_SOURCE_IS_CONTINUOUS,
174	.enable	= kvm_cs_enable,
175};
176EXPORT_SYMBOL_GPL(kvm_clock);
177
178static void kvm_register_clock(char *txt)
179{
180	struct pvclock_vsyscall_time_info *src = this_cpu_hvclock();
181	u64 pa;
 
 
 
 
182
183	if (!src)
184		return;
 
 
 
 
185
186	pa = slow_virt_to_phys(&src->pvti) | 0x01ULL;
187	wrmsrl(msr_kvm_system_time, pa);
188	pr_info("kvm-clock: cpu %d, msr %llx, %s", smp_processor_id(), pa, txt);
189}
190
191static void kvm_save_sched_clock_state(void)
192{
193}
194
195static void kvm_restore_sched_clock_state(void)
196{
197	kvm_register_clock("primary cpu clock, resume");
198}
199
200#ifdef CONFIG_X86_LOCAL_APIC
201static void kvm_setup_secondary_clock(void)
202{
203	kvm_register_clock("secondary cpu clock");
 
 
 
 
204}
205#endif
206
207/*
208 * After the clock is registered, the host will keep writing to the
209 * registered memory location. If the guest happens to shutdown, this memory
210 * won't be valid. In cases like kexec, in which you install a new kernel, this
211 * means a random memory location will be kept being written. So before any
212 * kind of shutdown from our side, we unregister the clock by writing anything
213 * that does not have the 'enable' bit set in the msr
214 */
215#ifdef CONFIG_KEXEC_CORE
216static void kvm_crash_shutdown(struct pt_regs *regs)
217{
218	native_write_msr(msr_kvm_system_time, 0, 0);
219	kvm_disable_steal_time();
220	native_machine_crash_shutdown(regs);
221}
222#endif
223
224static void kvm_shutdown(void)
225{
226	native_write_msr(msr_kvm_system_time, 0, 0);
227	kvm_disable_steal_time();
228	native_machine_shutdown();
229}
230
231static void __init kvmclock_init_mem(void)
232{
233	unsigned long ncpus;
234	unsigned int order;
235	struct page *p;
236	int r;
237
238	if (HVC_BOOT_ARRAY_SIZE >= num_possible_cpus())
239		return;
240
241	ncpus = num_possible_cpus() - HVC_BOOT_ARRAY_SIZE;
242	order = get_order(ncpus * sizeof(*hvclock_mem));
243
244	p = alloc_pages(GFP_KERNEL, order);
245	if (!p) {
246		pr_warn("%s: failed to alloc %d pages", __func__, (1U << order));
247		return;
248	}
249
250	hvclock_mem = page_address(p);
251
252	/*
253	 * hvclock is shared between the guest and the hypervisor, must
254	 * be mapped decrypted.
255	 */
256	if (sev_active()) {
257		r = set_memory_decrypted((unsigned long) hvclock_mem,
258					 1UL << order);
259		if (r) {
260			__free_pages(p, order);
261			hvclock_mem = NULL;
262			pr_warn("kvmclock: set_memory_decrypted() failed. Disabling\n");
263			return;
264		}
265	}
266
267	memset(hvclock_mem, 0, PAGE_SIZE << order);
268}
269
270static int __init kvm_setup_vsyscall_timeinfo(void)
271{
272#ifdef CONFIG_X86_64
 
 
273	u8 flags;
274
275	if (!per_cpu(hv_clock_per_cpu, 0) || !kvmclock_vsyscall)
276		return 0;
277
278	flags = pvclock_read_flags(&hv_clock_boot[0].pvti);
279	if (!(flags & PVCLOCK_TSC_STABLE_BIT))
280		return 0;
281
282	kvm_clock.vdso_clock_mode = VDSO_CLOCKMODE_PVCLOCK;
283#endif
284
285	kvmclock_init_mem();
286
287	return 0;
288}
289early_initcall(kvm_setup_vsyscall_timeinfo);
290
291static int kvmclock_setup_percpu(unsigned int cpu)
292{
293	struct pvclock_vsyscall_time_info *p = per_cpu(hv_clock_per_cpu, cpu);
294
295	/*
296	 * The per cpu area setup replicates CPU0 data to all cpu
297	 * pointers. So carefully check. CPU0 has been set up in init
298	 * already.
299	 */
300	if (!cpu || (p && p != per_cpu(hv_clock_per_cpu, 0)))
301		return 0;
302
303	/* Use the static page for the first CPUs, allocate otherwise */
304	if (cpu < HVC_BOOT_ARRAY_SIZE)
305		p = &hv_clock_boot[cpu];
306	else if (hvclock_mem)
307		p = hvclock_mem + cpu - HVC_BOOT_ARRAY_SIZE;
308	else
309		return -ENOMEM;
310
311	per_cpu(hv_clock_per_cpu, cpu) = p;
312	return p ? 0 : -ENOMEM;
313}
314
315void __init kvmclock_init(void)
316{
317	u8 flags;
318
319	if (!kvm_para_available() || !kvmclock)
320		return;
321
322	if (kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE2)) {
323		msr_kvm_system_time = MSR_KVM_SYSTEM_TIME_NEW;
324		msr_kvm_wall_clock = MSR_KVM_WALL_CLOCK_NEW;
325	} else if (!kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE)) {
 
 
 
 
 
 
 
326		return;
327	}
 
328
329	if (cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "kvmclock:setup_percpu",
330			      kvmclock_setup_percpu, NULL) < 0) {
 
331		return;
332	}
333
334	pr_info("kvm-clock: Using msrs %x and %x",
335		msr_kvm_system_time, msr_kvm_wall_clock);
336
337	this_cpu_write(hv_clock_per_cpu, &hv_clock_boot[0]);
338	kvm_register_clock("primary cpu clock");
339	pvclock_set_pvti_cpu0_va(hv_clock_boot);
340
341	if (kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE_STABLE_BIT))
342		pvclock_set_flags(PVCLOCK_TSC_STABLE_BIT);
343
344	flags = pvclock_read_flags(&hv_clock_boot[0].pvti);
 
 
 
345	kvm_sched_clock_init(flags & PVCLOCK_TSC_STABLE_BIT);
 
346
347	x86_platform.calibrate_tsc = kvm_get_tsc_khz;
348	x86_platform.calibrate_cpu = kvm_get_tsc_khz;
349	x86_platform.get_wallclock = kvm_get_wallclock;
350	x86_platform.set_wallclock = kvm_set_wallclock;
351#ifdef CONFIG_X86_LOCAL_APIC
352	x86_cpuinit.early_percpu_clock_init = kvm_setup_secondary_clock;
 
353#endif
354	x86_platform.save_sched_clock_state = kvm_save_sched_clock_state;
355	x86_platform.restore_sched_clock_state = kvm_restore_sched_clock_state;
356	machine_ops.shutdown  = kvm_shutdown;
357#ifdef CONFIG_KEXEC_CORE
358	machine_ops.crash_shutdown  = kvm_crash_shutdown;
359#endif
360	kvm_get_preset_lpj();
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
361
362	/*
363	 * X86_FEATURE_NONSTOP_TSC is TSC runs at constant rate
364	 * with P/T states and does not stop in deep C-states.
365	 *
366	 * Invariant TSC exposed by host means kvmclock is not necessary:
367	 * can use TSC as clocksource.
368	 *
369	 */
370	if (boot_cpu_has(X86_FEATURE_CONSTANT_TSC) &&
371	    boot_cpu_has(X86_FEATURE_NONSTOP_TSC) &&
372	    !check_tsc_unstable())
373		kvm_clock.rating = 299;
374
375	clocksource_register_hz(&kvm_clock, NSEC_PER_SEC);
376	pv_info.name = "KVM";
 
377}